80 likes | 88 Views
WG2 M achine learning for low frequency seismology. wg2-g2net@ego-gw.it. Alessandro Bertolini , Nikef , Netherland, WG2 leader Velimir Ili ć , MISASA, Serbia, WG2 co-leader. WG2 - Countries and institutes interested … so far.
E N D
WG2 Machine learning for low frequency seismology wg2-g2net@ego-gw.it Alessandro Bertolini, Nikef, Netherland, WG2 leader VelimirIlić, MISASA, Serbia, WG2 co-leader
WG2 - Countries and institutes interested … so far • Finland (University of Helsinki, AboAkademi University Turku) • France (Institut de Physique du Globe Paris) • Greece (Aristotle University of Thessaloniki) • Hungary (Wigner RCP Budapest) • Italy (ScuolaSuperiore S. Anna Pisa, INGV Pisa) • Malta (University of Malta) • Poland (University of Warsaw, Polish Academy of Sciences) • Serbia (Mathematical Institute SASA, Belgrade) • Netherlands (Nikhef, Amsterdam) • United Kingdom (University of Southampton) CA17137 1st conference – January 15th 2019 WG2 Session
WG2 focus – Newtonian noise on GW detectors dT Temperature fluctuations atmospheric Infrasound seismic Gravitational coupling between the mirrors and surrounding mass density fluctuations CA17137 1st conference – January 15th 2019 WG2 Session
WG2 focus – Newtonian noise on GW detectors • Challenges: • NN cannot be distinguished from GWs and cannot be shielded • Expectations about NN: • To be limitingthe reach of next generation detectors in the low frequency (2-20 Hz) band • To be observable in the existing instruments in bad weather conditions Current approach: Deploy arrays of environmental (seismic and pressure) sensors around the test masses to reconstruct the NN field Subtract its effect from the GW channel by means of an optimal (static) Wiener filter CA17137 1st conference – January 15th 2019 WG2 Session
WG2 objectives • Bringing the NN challenge to a new level by involving expertise outside GW physics Robotics for adaptive arrays ofenvironmental sensors • ML for seismic field modeling and NN field reconstruction: • wave propagation characteristics • noise stationarity and time evolution • design of optimized sensor arrays Complementary to WG3 CA17137 1st conference – January 15th 2019 WG2 Session
WG2 objectives • Interdiscpilinary research topics (check out MoU): Seismology: • earthquacke waveforms, seismic array intereferometry, Newtoian noise analysis Signal processing: • match-filtering, Wiener filtering, deconvolution Mathematical modeling: • Bayesian analysis, Markov chains, Fokker-Planck and Langevineequations Machine learning for robotics: • deep learning, reinforcment learning, Belief space planning Gravitational waves detection: • instrumentation, hardware and data processing CA17137 1st conference – January 15th 2019
Organization • WG2 tasks (preliminary): • Gravitational waves detectors instrumentation • Robotics • Seismology: hardware and data processing • Applied mathematics • …group is still being formed... • This week: • Task list will be finalized • Task leaders will be appointed Everyone interested in WG2 is invited to attend tomorrow morning group meeting and contribute to the open discussion CA17137 1st conference – January 15th 2019 WG2 Session
Today’s session • Tomasz Bulik (Warsaw University/Virgo) - Newtonian Noise subtraction test array at Virgo • Fabio Bonsignorio (Scuola Superiore Sant'Anna) - Robots for GW detectors and infrastructure monitoring and operation: Can Networks of Autonomous Robotics Vehicles help the characterization of Newtonian, Acoustic and Other Source Noise in GW detection? • JozinovićDario (INGV, Roma) - Machine Learning for seismic events • SoumenKoley (Nikhef, Netherland)-GW detector sites characterization with seismic arrays • Dr VelimirIlic (MISASA, Serbia) - Generalized maximum entropy inference for seismic models Enjoy! CA17137 1st conference – January 15th 2019 WG2 Session